Académique Documents
Professionnel Documents
Culture Documents
Abstract
Several deterministic and stochastic multi-variable global optimization algorithms (Conjugate Gradient, Nelder-Mead, Quasi-Newton, and Global) are
investigated in conjunction with energy minimization principle to resolve
the pressure and volumetric flow rate fields in single ducts and networks of
interconnected ducts. The algorithms are tested with seven types of fluid:
Newtonian, power law, Bingham, Herschel-Bulkley, Ellis, Ree-Eyring and
Casson. The results obtained from all those algorithms for all these types
of fluid agree very well with the analytically derived solutions as obtained
from the traditional methods which are based on the conservation principles
and fluid constitutive relations. The results confirm and generalize the findings of our previous investigations that the energy minimization principle is
at the heart of the flow dynamics systems. The investigation also enriches
the methods of Computational Fluid Dynamics for solving the flow fields
in tubes and networks for various types of Newtonian and non-Newtonian
fluids.
Keywords: energy minimization; fluid dynamics; global multi-variable optimization; pressure; flow rate; tube; network; deterministic, stochastic; Conjugate Gradient; Nelder-Mead; Quasi-Newton; Global; Newtonian; power
law; Bingham; Herschel-Bulkley; Ellis; Ree-Eyring; Casson.
Introduction
The traditional method for resolving the pressure and volumetric flow rate fields in
fluid conducting devices is to use the conservation principles, which are normally
based on the mass continuity and momentum conservation, in conjunction with
the constitutive relations that link the stress to the rate of deformation and are
specific to the particular types of fluid employed to model the flow [14]. For
single conduits, this usually results in an analytical expression that correlates the
volumetric flow rate to the applied pressure drop as well as other dependencies on
the parameters of the conduits, such as the radius and length of the tube, and the
parameters of the fluid such as the shear viscosity and yield stress. For networks
of interconnected conduits, the analytical expression for the single conduit for the
particular fluid model can be exploited in a numeric solution scheme, which is
normally of iterative nature such as the widely used Newton-Raphson procedure
for solving a system of simultaneous non-linear equations, in conjunction with the
mass conservation principle and the given boundary conditions to obtain the flow
fields in the network.
Recently the energy minimization principle in the flow through single conduits
and networks of interconnected conduits was investigated [5, 6] as a possible underlying rule for the flow phenomena that can be exploited to resolve the pressure
and volumetric flow rate fields. While in [5] the issue was investigated numerically
in relation to the flow of Newtonian fluids using a stochastic simulated annealing
[79] procedure, in [6] it was investigated analytically in relation to the flow of
Newtonian and power law fluids using standard analytical optimization methods
from Calculus.
In the present study we continue those investigations but this time the issue is investigated numerically in relation to the flow of Newtonian and six nonNewtonian fluid models using three deterministic and one stochastic global multi2
variable optimization algorithms. The six non-Newtonian fluid models are: power
law, Bingham, Herschel-Bulkley, Ellis, Ree-Eyring and Casson. The three employed deterministic algorithms are: Conjugate Gradient, Nelder-Mead, and QuasiNewton, while the stochastic algorithm is the Stochastic Global. Several types
of network, which include one-dimensional (1D) two-dimensional (2D) and threedimensional (3D), of different geometries and topologies, such as fractals and irregulars based on cubic and orthorhombic lattices, are used to examine and validate
the energy minimization proposal.
All the results obtained in the current study support the generalization of the
energy optimization as a fundamental principle that underlies the flow phenomena
in the Newtonian and non-Newtonian fluid dynamics systems. The study also adds
more tools to the Computational Fluid Dynamics as these computational methods,
which are based on energy minimization, can be used for finding the pressure and
flow rate fields in tubes and networks.
The plan for this paper is as follow: in the next section 2 we present a
general theoretical background about the energy minimization principle and its
use in association with the global multi-variable optimization algorithms to resolve
the flow fields in tubes and networks of interconnected tubes for different types
of fluid. This will be followed in section 3 by discussing the implementation of
the energy minimization principle within global multi-variable optimization codes
and the results that have been obtained from different optimization algorithms
using various types of fluid and different kinds of network where we analyze the
results and compare them to the standard analytical solutions as obtained from and
verified by the traditional methods which are based on the conservation principles
and constitutive relations for solving the flow fields. Finally, in section 4 the
paper is concluded by outlining the main issues that have been examined in this
study and their theoretical and practical significance.
Theoretical Background
In this investigation, we assume an incompressible, laminar, pressure-driven, fullydeveloped flow with minimal entry and exit effects and negligible viscous frictional
losses. We also assume minor effects from external body forces such as gravitational
attraction and electromagnetic interaction. The single conduits, as well as the
conduits in the interconnected networks, are assumed to be rigid of uniform and
circularly-shaped cross sections along their axial dimension.
As for the boundary conditions, we assume Dirichlet-type pressure boundary
conditions. The last assumption is imposed only for convenience and practical
considerations; otherwise the energy minimization argument, when established,
will not be restricted to such conditions which basically reflect the way used to
model and portray the flow system by the observer and hence the type of the
boundary conditions does not represent an inherent characteristic of the system
that is due to determine its final outcome.
Concerning the type of fluid, we assume a generalized Newtonian fluid which
in this investigation includes Newtonian, Ostwald-de Waele, Bingham, HerschelBulkley, Ellis, Ree-Eyring and Casson models. The constitutive relations for these
models, as well as the analytical expressions for their volumetric flow rate through
rigid uniform pipes of circular cross sections, are given in Table 1.
The time rate of energy consumption, I, for transporting a certain amount of
fluid through a single conducting device, considering the pre-stated flow assumptions, is given by
I = p Q
(1)
where p is the pressure drop across the conducting device and Q is the volumetric
flow rate of the transported fluid through the device. For a flow conducting device
It (p1 , . . . , pN ) =
m
X
pl Ql
(2)
l=1
where N is the number of the boundary and internal nodes. For a single duct,
the conducting elements are the discretized sections, while for a network they
represent the conducting ducts as well as their discretized sections if discretization
is employed.
Starting from randomly selected values for the internal nodal pressure, with the
given pressure values for the inlet and outlet boundary nodes, the role of the global
multi-variable optimization algorithms in the above-described energy consumption
model is to minimize the cost function, which is the time rate of the total energy
consumption for fluid transportation It as given by Equation 2, by varying the
values of the internal nodal pressure while holding the pressure values at the inlet
and outlet boundary nodes as constants. The volumetric flow rates, Q, that have
to be used in Equation 2 for the employed fluid models are given by the expressions
in Table 1.
Table 1: The constitutive relations and the volumetric flow rates, Q, for the seven
fluid models used in this investigation. These volumetric flow rates are derived for
rigid uniform pipes of circular cross sections. The meanings of the symbols are
given in Nomenclature 5.
Model
Newtonian
Constitutive
=
Power Law
= k n
Bingham
Q
R4 p
8Lq
R4
8L
= C + 0
Ree-Eyring
= C n + 0
"
1 #
= 0 1 +
1/2
= c arcsinh 0c
Casson
1/2 = (K)
Herschel-Bulkley
Ellis
p
k
4n
3n+1
4
R4 p 1
0
8LC
3 w
1/2
n
C
L
p
3
"
3
R w
40
R3
3K
w
1/2
2L 11/n
R
4
3
0
w
+1
h
(w 0 )
(w 0 )
3+1/n +
#
1
1+1/n
4
+3
20 (w 0 )
2+1/n
02
1+1/n
w
1/2
i
c w2 + 2c3 cosh wc 2c2 w sinh wc 2c3
4
7/2
4 0 w
3
w
+ 03 w
4
7
R3 c
3
w
0
+ 0
1+
execution.
The size of the networks used in the investigation ranges between a small number to several hundreds, and even thousands in some cases, of nodes and segments.
The multi-dimensional networks used in this investigation are of two main types:
two-dimensional of fractal and rectangular morphology, and three-dimensional built
on cubic and orthorhombic lattice structures. The fractals are based on fractal
branching patterns where each generation of the branching tubes in the network
has a specific number of branches related to the number of branches in the parent
generation, such as 3:1, as well as specific branching angle, radius branching ratio
and length to radius ratio. The cubic and orthorhombic networks are based on cubic
and orthorhombic three-dimensional lattice structures respectively where the radii
of the tubes in the network are subjected to random statistical distributions such as
the uniform or the normal distributions. Similar statistical distributions were also
applied to the two-dimensional rectangular networks. A graphic demonstration
of three main types of network; namely one-dimensional linear, two-dimensional
fractal and three-dimensional orthorhombic; is given in Figure 1.
The size of the difference between the numerical optimization solutions and
the analytical solutions depends mainly on the particular algorithm, the type and
parameters of the fluid and the size and type (1D, 2D or 3D and fractal or orthorhombic) of the network. A typical size of the average percentage relative difference between the numerical optimization solutions and the analytical solutions is
less than 0.5% for the one-dimensional networks, about 1% for the two-dimensional
networks, and 2-3% for the three-dimensional networks. In most cases, the best
optimization algorithm with regard to the agreement of its solution with the analytical solution is the Global while the worst is the Nelder-Mead. The latter has
also failed to converge in some cases.
In our view, the observed discrepancy between the numerical optimization so-
lutions and the analytical solutions in all the investigated cases can be justified
by premature convergence of the optimization algorithms due to practical limits
on their convergence criteria as well as numerical errors arising from limitations of
the employed optimization algorithms and codes plus non-linearities, especially in
some cases of non-Newtonian models with extreme non-linear characteristics such
as high shear thinning and yield stress.
There are many computational issues related to the performance and convergence behavior of these algorithms and their relation to the type and parameters
of the fluids and the size and type of the networks. However, these technical details are irrelevant to the current study whose main objective is to provide further
validation and demonstration for the use of energy minimization principle in resolving the flow fields in tubes and networks, rather than investigating numerical
and computational issues.
Table 2: A sample of the one-dimensional linear networks of serially connected
rigid uniform tubes of circular cross sections with the given number of segments
(NS) that have been used in the current investigation to compare the analytical
solutions to the solutions of the global optimization algorithms.
Network
1
2
3
4
NS
7
8
8
6
Lengths (cm)
80,60,70,90,90,50,60
4.8,4,4.8,4,5.6,7.2,5.6,4
2.4,2,2.4,2,2.8,3.6,2.8,2
14,8,8,14,14,17
Radii (cm)
2.5,2.1,1.8,1.3,1.7,2.6,1.6
1,1.3,0.75,0.6,0.5,0.85,1.1,1.2
0.6,0.5,0.44,0.28,0.38,0.49,0.57,0.51
2.4,2,1.76,2.232,1.52,1.8
3000
3000
Analytical
CG
2500
2000
2000
1500
1500
1000
1000
500
500
0
0
Analytical
NM
2500
0
0
3000
2000
1500
1500
1000
1000
500
500
1
Analytical
SG
2500
2000
0
0
(b) Nelder-Mead.
Analytical
QN
2500
0
0
(c) Quasi-Newton.
Figure 2: Comparison between the analytical solution and the solutions obtained
from the indicated global optimization algorithms which are based on the energy
minimization principle for a shear thickening power law fluid with n = 1.2 and k =
0.05 Pa.sn . The computations were carried out using the first network of Table 2
with inlet and outlet pressure boundary conditions of 3000 Pa and 0 Pa respectively.
The volumetric flow rate through the network is Q = 1.690 104 m3 .s1 . In all
four sub-figures, the vertical axis represents the network axial pressure in Pa while
the horizontal axis represents the network axial coordinate in m. Similar results
were obtained for the Newtonian model which is a special case of the power law
model with n = 1.
10
2000
2000
Analytical
CG
1500
1500
1000
1000
500
500
0
0
0.1
0.2
0.3
0
0
0.4
1500
1000
1000
500
500
0.2
0.3
0.2
2000
Analytical
QN
0.1
0.1
0.3
0.4
(b) Nelder-Mead.
1500
0
0
Analytical
NM
0
0
0.4
(c) Quasi-Newton.
Analytical
SG
0.1
0.2
0.3
0.4
Figure 3: Comparison between the analytical solution and the solutions obtained
from the indicated global optimization algorithms which are based on the energy
minimization principle for a shear thinning yield stress Herschel-Bulkley fluid with
n = 0.6, C = 0.008 Pa.sn and 0 = 1.0 Pa. The computations were carried out using
the second network of Table 2 with inlet and outlet pressure boundary conditions
of 2000 Pa and 0 Pa respectively. The volumetric flow rate through the network
is Q = 1.022 101 m3 .s1 . In all four sub-figures, the vertical axis represents the
network axial pressure in Pa while the horizontal axis represents the network axial
coordinate in m. Similar results were obtained for the Bingham model which is a
special case of the Herschel-Bulkley model with n = 1.
11
1000
1000
Analytical
CG
800
800
600
600
400
400
200
200
0
0
0.05
0.1
0.15
0
0
0.2
400
400
200
200
0.1
0.15
0.15
0.2
Analytical
SG
800
600
0.05
0.1
1000
600
0
0
0.05
(b) Nelder-Mead.
Analytical
QN
800
Analytical
NM
0
0
0.2
(c) Quasi-Newton.
0.05
0.1
0.15
0.2
Figure 4: Comparison between the analytical solution and the solutions obtained
from the indicated global optimization algorithms which are based on the energy
minimization principle for an Ellis fluid with 0 = 0.18 Pa.s, = 2.4 and 1/2 =
1025 Pa. The computations were carried out using the third network of Table 2
with inlet and outlet pressure boundary conditions of 1000 Pa and 0 Pa respectively.
The volumetric flow rate through the network is Q = 3.160 106 m3 .s1 . In all
four sub-figures, the vertical axis represents the network axial pressure in Pa while
the horizontal axis represents the network axial coordinate in m.
12
1500
Analytical
CG
1500
1000
1000
500
500
0
0
0.2
0.4
0
0
0.6
Analytical
QN
500
500
0.4
0.4
1500
1000
0.2
0.2
0.6
(b) Nelder-Mead.
1000
0
0
Analytical
NM
0
0
0.6
(c) Quasi-Newton.
Analytical
SG
0.2
0.4
0.6
Figure 5: Comparison between the analytical solution and the solutions obtained
from the indicated global optimization algorithms which are based on the energy
minimization principle for a Ree-Eyring fluid with 0 = 0.018 Pa.s and c = 300 Pa.
The computations were carried out using the fourth network of Table 2 with inlet
and outlet pressure boundary conditions of 1500 Pa and 0 Pa respectively. The
volumetric flow rate through the network is Q = 4.991 103 m3 .s1 . In all four
sub-figures, the vertical axis represents the network axial pressure in Pa while the
horizontal axis represents the network axial coordinate in m.
13
2500
Analytical
CG
2000
2500
2000
1500
1500
1000
1000
500
500
0
0
0.2
0.4
0
0
0.6
1000
1000
500
500
0.4
0.6
Analytical
SG
2000
1500
0.2
0.4
2500
1500
0
0
0.2
(b) Nelder-Mead.
Analytical
QN
2000
Analytical
NM
0
0
0.6
(c) Quasi-Newton.
0.2
0.4
0.6
Figure 6: Comparison between the analytical solution and the solutions obtained
from the indicated global optimization algorithms which are based on the energy
minimization principle for a Casson fluid with 0 = 0.01 Pa.s and 0 = 1.0 Pa.
The computations were carried out using the fourth network of Table 2 with inlet
and outlet pressure boundary conditions of 2500 Pa and 0 Pa respectively. The
volumetric flow rate through the network is Q = 9.627 103 m3 .s1 . In all four
sub-figures, the vertical axis represents the network axial pressure in Pa while the
horizontal axis represents the network axial coordinate in m.
14
Conclusions
15
Nomenclature
Newtonian viscosity
shear stress
1/2
It
tube length
pressure
tube radius
0
2
in Ellis model
Rp
)
2L
16
References
[1] A.H.P. Skelland. Non-Newtonian Flow and Heat Transfer. John Wiley and
Sons Inc., 1967. 2
[2] R.B. Bird; R.C. Armstrong; O. Hassager. Dynamics of Polymeric Liquids,
volume 1. John Wiley & Sons, second edition, 1987. 2
[3] F.M. White. Viscous Fluid Flow. McGraw Hill Inc., second edition, 1991. 2
[4] T.C. Papanastasiou; G.C. Georgiou; A.N. Alexandrou. Viscous Fluid Flow.
CRC Press, first edition, 1999. 2
[5] T. Sochi.
Submitted, 2014.
arXiv:1408.0357. 2, 15
[6] T. Sochi. Energy minimization for the flow in ducts and networks. Submitted,
2014. arXiv:1412.1804. 2, 15
[7] N. Metropolis; A.W. Rosenbluth; M.N. Rosenbluth; A.H. Teller; E. Teller.
Equation of State Calculations by Fast Computing Machines. Journal of
Chemical Physics, 21:10871092, 1953. 2
[8] S. Kirkpatrick; C.D. Gelatt Jr.; M.P. Vecchi. Optimization by Simulated
Annealing. Science, 220(4598):671680, 1983. 2
y. Thermodynamical approach to the traveling salesman problem: An
[9] V. Cern
efficient simulation algorithm. Journal of Optimization Theory and Applications, 45(1):4151, 1985. 2
[10] W.H. Press; S.A. Teukolsky; W.T. Vetterling; B.P. Flannery.
Numerical
[11] C.G.E. Boender; A.H.G. Rinnooy Kan; G.T. Timmer; L. Stougie. A stochastic
method for global optimization. Mathematical Programming, 22(1):125140,
1982. 6
18